Precomputed Atmospheric Scattering:
a New Implementation

Eric Bruneton, 2017


This document presents a new implementation of our Precomputed Atmospheric Scattering paper. This new implementation is motivated by the fact that the original implementation:

To address these concerns, our new implementation: In addition, the new implementation adds support for the ozone layer, and for custom density profiles for air molecules and aerosols.

The sections below explain how this new implementation can be used, present its structure and its documentation and give more details about its tests.


Our new implementation can be used in C++ / OpenGL applications as explained in model.h, and as demonstrated in the demo in atmosphere/demo. To run this demo, simply type make demo in the main directory. A WebGL2 version of this demo is also available online.

The default settings of this demo use the real solar spectrum, with an ozone layer. To simulate the settings of the original implementation, set the solar spectrum to "constant", and turn off the ozone layer.


The source code is organized as follows:

The most important files are the 5 files in the atmosphere directory. They contain the GLSL shaders that implement our atmosphere model, and provide a C++ API to precompute the atmosphere textures and to use them in an OpenGL application. This code does not depend on the content of the other directories, and is the only piece which is needed in order to use our atmosphere model on GPU.

The other directories provide examples and tests:


The documentation consists of a set of web pages, generated from the extensive comments in each source code file:


To reduce the risk of implementation errors, two kinds of verifications are performed:

The main issue to implement this is that a GLSL compiler cannot check the dimensional homogeneity, unlike a C++ compiler (see for instance Boost.Units). Our solution to this problem is to write our GLSL code in such a way that it can be compiled both by a GLSL compiler and by a C++ compiler. For this:

Thanks to this double GLSL and C++ compilation, the unit tests for the GLSL code can then be implemented either in GLSL or in C++. We chose C++ because it is much more practical. Indeed, a C++ unit test does not need to send data to the GPU and to read back the test result, unlike a GLSL unit test.